A Dissertation Submitted in Partial Fulfillment of the Requirements for the Degree of Master’s in Information and Communication Science and Engineering of the Nelson Mandela African Institution of Science and TechnologyIn Tanzania, smallholder farmers are mainly involved in farming activities which contribute significantly to food security and nutrition. Kagera, Mbeya and Arusha regions lead in high banana production, however, diseases and pests are threats to the yields. Early detection and identification of banana diseases is still a challenge for smallholder farmers and extension officers due to lack of the necessary tools such as sensors and mobile applications. In this research, an early detection tool for banana fungal diseases was developed. This research presents deep learning models trained for the detection of banana diseases of Fusarium wilt and Black sigatoka and deployed on a smartphone for the early detection of the diseases in real time. The five models selected and trained include VGG16, Resnet18, Resnet50, Resnet152 and InceptionV3. The VGG16 model achieved an accuracy of 97.26%, Resnet50 achieved an accuracy of 98.8%, Resnet18 achieved an accuracy of 98.4%, InceptionV3 achieved an accuracy of 95.41% and Resnet152 achieved an accuracy of 99.2%. We therefore, used InceptionV3 model for deployment in mobile phones because it has low computation cost and low memory requirements of the all models. The developed tool was capable of detecting diseases with 99% of confidence of the captured leaves from the real world environment. The system was developed on Android based application in English language. The developed tool has the potential to support smallholder farmers and extension officers to detect banana fungal disease at early stages. We conclude that early detection of the diseases is important. Hence control and management of banana fungal diseases will be done early for the improvement of banana yield.